YOLO FAMILY ARCHITECTURES FOR EXPLOSIVE OBJECT DETECTION IN EMBEDDED SYSTEMS OF AUTONOMOUS UNMANNED VEHICLES: A COMPARATIVE ANALYSIS

Abstract

Relevance. The need to choose an effective YOLO architecture for explosive object detection in unmanned aerial vehicle embedded systems is driven by limited computational resources, real-time constraints, and energy efficiency requirements. The object of research is explosive object detection in embedded autonomous unmanned vehicle systems using YOLO family architectures. The subject of the research is the comparative efficiency of YOLO models in terms of detection accuracy, throughput, model size, and energy consumption on embedded hardware platforms. The purpose of this paper is to develop and apply an integral efficiency criterion based on neural network methods for selecting the most suitable YOLO architecture for explosive object detection and humanitarian demining tasks. Research results. The proposed criterion was applied to six YOLO models (v5n, v8n, v8s, v10n, v11n, v11s) deployed on Raspberry Pi 3 Model B+ and Raspberry Pi 4 Model B with 4 GB of RAM. Comparative evaluation under three scenarios showed substantial differences in detection quality, throughput, model size, and energy consumption. The ranking confirms the practical value of the methodology for model selection in embedded systems. Conclusions. The methodology provides a transparent and reproducible basis for comparing YOLO architectures in resource-constrained environments. Real-time operation remains challenging on low-power platforms, which justifies further optimization of deployment and inference.